基于多智能体深度强化学习的配-微网协同优化调度研究

Collaborative Optimization Scheduling of Distribution Network and Microgrids Based on Multi Agent Deep Reinforcement Learning

近年来,微电网在新型电力系统建设中扮演着愈发重要的角色。但分布式灵活资源大量接入,调度中心面临的通信和计算任务日益繁重,集中式调度策略难以有效兼顾多主体的利益诉求并保证计算高效性。为此,提出了一种基于多智能体深度强化学习算法(MADRL)的配-微网协同优化调度策略。首先,针对含多类型资源集群的协同调度问题,以最小化微电网群总运行成本为目标进行建模;其次,设计了一个集中训练和分散执行的多智能体深度强化学习框架,将协同调度问题转化为马尔可夫博弈,并改进含熵约束的算法,在策略训练阶段使用全局状态信息作为输入,训练完成后每个智能体由局部观测执行优化决策;最后,在修改后的配电网系统开展了算例分析,验证了所提方法的有效性。

In recent years,microgrids have played an increasingly important role in the construction of new type power systems.However,with the massive access of distributed flexible resources,the communication and computing tasks faced by the dispatch center are becoming increasingly heavy.Centralized scheduling strategies are difficult to effectively balance the interests and demands of multiple parties and ensure computational efficiency.The collaborative optimization scheduling strategy for microgrid is proposed based on multi-agent deep reinforcement learning algorithm.Firstly,aiming at the collaborative scheduling problem with multiple types of resource clusters,the model is set up with the objective function of minimizing the operating cost of microgrids;Secondly,a multi-agent deep reinforcement learning framework with centralized training and decentralized execution is designed to transform the collaborative scheduling problem into Markov game,the algorithm with entropy constraints is improved.During the strategy training phase,global state information is used as input,each agent makes optimization decisions through local observations after training;Finally,a case study is conducted based on the modified distribution network system to verify the effectiveness of the proposed method.